Prediksi Freezing Of Gait (FoG) Pada Penyakit Parkinson Menggunakan Data Multichannel Akselerometer

Casey, Michelle (2025) Prediksi Freezing Of Gait (FoG) Pada Penyakit Parkinson Menggunakan Data Multichannel Akselerometer. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023211003-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (6MB) | Request a copy

Abstract

Penyakit Parkinson merupakan gangguan neurodegeneratif progresif yang ditandai oleh degenerasi neuron dopaminergik pada substantia nigra pars compacta (SNpc) di basal ganglia. Salah satu gejala umum pada pasien Parkinson adalah Freezing of Gait (FoG), yaitu gangguan gaya berjalan di mana pasien kehilangan kemampuan untuk melangkah meskipun memiliki intensi untuk berjalan. FoG meningkatkan risiko jatuh, sehingga diperlukan sistem prediksi yang cepat dan akurat untuk mengidentifikasi kejadian FoG. Namun, penelitian terkait prediksi FoG masih terbatas karena mekanisme fisiologis yang mendasari fenomena ini belum sepenuhnya dipahami. Karakteristik FoG yang muncul secara tiba-tiba, bersifat sementara, serta bervariasi antar individu menjadikannya tantangan prediksi yang kompleks. Penelitian sebelumnya umumnya berfokus pada ekstraksi fitur manual atau fitur yang diperoleh melalui deep learning. Padahal, kombinasi kedua pendekatan ini berpotensi memberikan pemahaman yang lebih komprehensif terhadap fenomena FoG. Oleh karena itu, penelitian ini mengembangkan sistem prediksi FoG dengan menggabungkan fitur manual dengan fitur yang dihasilkan melalui deep learning. Fitur manual yang digunakan meliputi fitur temporal gait, seperti cadence, stride duration, dan stance phase percentage, serta fitur domain frekuensi, seperti dominant frequency, freeze index, dan gait variability. Untuk meningkatkan performa sistem, penelitian ini juga menerapkan metode personalized labeling berdasarkan fitur yang memiliki korelasi tinggi dengan FoG, guna mendefinisikan fase pre-FoG dengan mempertimbangkan variasi durasi FoG antar individu. Selanjutnya, proses klasifikasi pola gait dilakukan menggunakan jaringan berbasis Transformer yang mampu menangkap hubungan antar saluran akselerometer secara efektif, dengan total target empat kelas yang meliputi berjalan normal, pre-FoG, FoG, dan post-FoG. Berdasarkan hasil pengujian yang telah dilakukan, diperoleh hasil bahwa kombinasi fitur dan personalized labeling memberikan peningkatan F1-Score sebesar 26,77% dan 28,25% masing-masing untuk kelas pre-FoG dan post-FoG. Sehingga, dapat disimpulkan bahwa penambahan fitur-fitur temporal dan frekuensi mampu meningkatkan kemampuan model dalam membedakan fase transisi seperti pre-FoG dan post-FoG, yang sebelumnya sulit dikenali hanya dengan menggunakan sinyal akselerometer murni.
=========================================================================================================================================
Parkinson’s disease is a progressive neurodegenerative disorder characterized by the degeneration of dopaminergic neurons in the substantia nigra pars compacta (SNpc) of the basal ganglia. One of its common motor symptoms is Freezing of Gait (FoG), a gait disturbance in which patients temporarily lose the ability to initiate movement despite having the intention to walk. FoG significantly increases the risk of falls, necessitating the development of fast and accurate prediction systems. However, research on FoG prediction remains limited due to the incomplete understanding of its underlying physiological mechanisms. The unpredictable, transient, and highly individual-specific nature of FoG presents a complex challenge for prediction models. While prior studies have predominantly relied on either handcrafted features or deep learning-based features, combining both approaches has the potential to yield a more comprehensive understanding of FoG. This study proposes a FoG prediction system that integrates handcrafted features with deep learning representations. The handcrafted features include temporal gait parameters such as cadence, stride duration, and stance phase percentage, as well as frequency-domain features such as dominant frequency, freeze index, and gait variability. To further improve system performance, a personalized labeling strategy is introduced, which defines the pre-FoG phase based on features with high correlation to FoG events, thereby accommodating individual variability in FoG duration. Gait pattern classification is performed using a Transformer-based network capable of effectively capturing inter-channel relationships in multi-channel accelerometer signals. The system classifies gait into four classes: normal walking, pre-FoG, FoG, and post-FoG. Experimental results demonstrate that the combination of feature integration and personalized labeling improves the F1-Score by 26.77% and 28.25% for the pre-FoG and post-FoG classes, respectively. These findings suggest that incorporating temporal and frequency features enhances the model’s ability to distinguish transitional gait phases that are difficult to detect using raw accelerometer signals alone.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Penyakit Parkinson, Freezing of Gait, Pola Gait, Personalized Labelling, Transformer, Parkinson’s Disease, Freezing of Gait, Gait Pattern, Personalized Labelling, Transformer
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Michelle Casey
Date Deposited:	04 Aug 2025 07:16
Last Modified:	04 Aug 2025 07:16
URI:	http://repository.its.ac.id/id/eprint/126516

Actions (login required)

View Item